SMART MONEY INDICATORS
CHAPTER 25 THE SECRETS OF
SUCCESSFUL TRADING
by Fernando Diz'1
A question that is often asked by anyone aspiring to be a trader is: What makes a trader successful?
Anyone who has read Jack Schwager's two Market Wizards books will know that there are at least three components to successful trading: the trader's psychological makeup, the trading system's edge, and strict money management. Although knowing about these provide you with very useful general guidelines, any trader aspiring to be successful could benefit from a more systematic study of traders' successes and, more important, failures. With this purpose in mind, I conducted a study of 925 CTA programs over the 1974-1995 period2. Since I am not a psychologist, I only focused on the last two components of successful trading: the edge and money management. The purpose of the study was to find out how the edge and money management affect a program's success.3
1 Fernando Diz is an assistant professor of finance at Syracuse University School of Management
2 The data used in study was provided by the Barclay Trading Group. Special thanks to Sol Waksman and Peter Nicols for their helpful comments.
3 Diz, Fernando (1995), CTA Survival and Return Distribution Characteristics, working paper, Syracuse University.
The definition of success in this study was very simple: a CTA or program was successful if it was able to stay in business. Of the 925 CTAs or programs studied between 1974 and 1995, 490 are still in business while the remaining 435 went out of business.
The "edge" of a program was proxied by five different and complementary measures:
• Monthly compounded return-bottom line performance of a program or trader.
• Maximum monthly return-return on the most successful month. Proxy for the program's return potential and whether it is trading at optimal levels.
• Sharpe Ratio-measure of returns per unit of standard deviation. Allows you compare returns of programs with different volatility.
• Skewness of returns- a positive number tells you that the program has a tendency to generate returns higher than the mean more often than normal.
• Kurtosis of returns-measures how returns are packed around the mean relative to a normal distribution. A positive number tells you that there is higher probability of finding returns near the mean compared to a normal distribution.
The first one is the monthly compounded rate of return over the life of the program.
This measures the level of returns that the program has been able to yield over its life. To give you an idea of the range of returns that you can find among different programs over their lives, the maximum monthly return was 10.36 percent and the minimum monthly return was -6.04 percent.
The second measure of the edge was the return on the most successful month. This measure focuses on the magnitude of
potential gains and gives you an idea of the magnitude of the program's edge and whether the program is trading at optimal levels or not. The program with the most successful month had a return of 319 percent while the program with the smallest most successful month had a return of 0.8 percent.
197
The third measure used was the Sharpe ratio. This is a measure of returns per unit of risk. It allows you to compare returns from programs with very different risk characteristics. Suppose that you have two programs A and B that have the same monthly compounded return, but program A has a Sharpe ratio that is two times as large as that of program B. What this means is that program A can generate the same returns as program B but with half the risk or volatility. The last two measures of the edge, skewness and kurtosis, are rather technical and give you an idea of how capable a program has been in generating larger returns with a high degree of probability.
just as edge measures increases in a CTMs equity curve, money management looks at the dips and recoveries in that curve.
The quality of the program's money management was proxied by five different measures:
• Standard deviation of returns-consistency with which programs achieve their returns. Gives you an idea of what kind of drawdowns and large returns to expect.
• Maximum monthly drawdown-magnitude of potential loses. Gives you an idea of risk controls and whether the program trades at optimal levels or not.
• Months to recover from worst drawdown-other things equal, programs with good money management will recover quicker than programs with poor money management.
• Standard deviation of times to recover from drawdown-measures the consistency with which programs come out of drawdowns.
• Time to recover from worst drawdown as a percent of the program's life indicates how much of a program's business life was spent recovering from the worst drawdown.
The first one was the standard deviation of monthly returns. The standard deviation gives you an idea of the probability of big hits or large drawdowns relative to the program's average monthly return. The second
198
measure of money management was the maximum monthly drawdown. This measure focuses on the magnitude of potential losses and gives you an idea of both risk controls and whether the program is trading at optimal levels or not. To give you a better idea of the magnitude of drawdowns that you can find, the largest drawdown was 81 percent, while the smallest drawdown was no drawdown at all.
The average monthly drawdown for all programs studied was 16 percent. Although looking at any single monthly drawdown is important, perhaps even more important is the amount of time that it takes a program to come out of a loss. For example, suppose that a program's initial equity is $1,000,000 and that in the next month the program experiences a 25 percent drawdown, reducing its equity to $750,000. What is the maximum number of months that would take the program to bring the equity back to $1,000,000? This is what the third money management variable measures. It took 137 months for the worst program in this category to recover from its worst drawdown. It is not surprising that after failing to recover from its worst drawdown in more than 11 years, this program went out of business. It took an average of 20 months for the average program to recover from its worst drawdown. This measure is important because it can be theoretically shown that a program with good money management can recover from drawdowns much faster than a program with poor money management. What does this mean? It means that if you look at two traders that have the same edge (system), the one with better money management will recover from drawdowns faster. What if one trader has a larger edge than the other?
If both traders use good money management, the one with the larger edge is likely to recover faster from drawdowns. So much for the academic explanation. What is more likely to happen out there?
I have looked at the most successful trend-follower traders out there and found that their distribution of monthly returns is remarkably similar; they have a comparable edge. The big difference between them is in their losses. The traders with the highest average returns have consistently smaller losses. This is an example of money management improving your existing edge.
The fourth money management variable measures the consistency with which programs recover from drawdowns. A large number for this measure means that programs do not recover consistently from their losses over time. This is probably related to both the edge of the program and its money management alike.
199
The last money management variable measures the longest time that a program spent recovering from a drawdown expressed as a fraction of the program's life. For example, if a program had been in business for 24 months and it spent 12 months recovering from its worst drawdown the value of this variable would be 0.5. This variable tries to capture the importance of a program's worst drawdown relative to the time that it has been in business.
The results of the study show that both the edge and money management strongly affect CTAs likelihood of success. This general result only confirms what The Market Wizards had suggested. But my interest was much more specific than that. I wanted to know if there were important differences between the successful (survivors) and the unsuccessful (out of business) CTAs or programs. First, I looked at their edge. You can see the results in Exhibit 25.1.
Two things definitely stand out from looking at the results in Exhibit 25.1.
Successful programs were able to yield a monthly compounded rate of return 48 percent higher than unsuccessful programs.
EXHIBIT 25.1 Average
Values for the "Edge" Variables for Successful and Unsuccessful Programs
More importantly, not only were they able to generate larger returns for their clients but they did so with lower risk than the unsuccessful programs. The average Sharpe ratio for successful programs was 61 percent larger than the one for unsuccessful programs. The ability to generate larger returns with lower risk leads us to the second component of successful trading: money management. The results of a similar comparison are shown in Exhibit 25.2.
200
EXHIBIT 25.2 Average Values for Money Management Variables for Successful and Unsuccessful Programs
The results in Exhibit 25.2 are quite revealing. Successful programs returns are 11 percent less variable than those of their unsuccessful counterparts. Furthermore, successful traders experience lower maximum drawdowns, they recover from drawdowns faster, and they are more consistent in doing so. Finally, notice that unsuccessful programs spend a larger proportion of their business life recovering from their worst drawdown!
The lesson from these results is clear: successful traders have a larger edge and much better money management than unsuccessful traders. Remember that these are average differences between successful and unsuccessful programs. These results do not tell you however, which one of these two components might be more important in explaining success. This is exactly the question that I wanted to answer. The question that I asked myself was: If I had to explain success with only one, or two variables, which ones would give me the most explanatory power for discriminating between successful and unsuccessful traders? As you will see, the results were rather astounding.
If I had to explain success with only one variable, the one that predicted success or failure most accurately was the amount of time a program spent recovering from its worst drawdown as a percentage of its life.
201
Very small values of this variable were associated with success, and very large values predicted failure. An important lesson to be drawn from this finding that is particularly suited to traders who have been in business for a very short time is:
Avoid large drawdowns and recover from them quickly if you want to stay in business (but do not increase risk to do it!). More interestingly, when you include the maximum number of months that it takes a trader to recover from its worst drawdown in the analysis, these two variables alone explained 88 percent of the predictive power of a model with all the edge and money management variables put together! This clearly demonstrates that although unsuccessful traders have a smaller edge over successful ones, their smaller edge is not the cause of their failure;
their money management is! In fact, their smaller edge may very well be a result of poor money management. This is quite revealing. What it basically tells us is that most CTAs or programs that have gone out of business had enough of an edge to stay in business! The cause of their failure can be traced to poor money management.
Two of the most important components of successful trading are the edge of a system and sound money management. This study confirms this notion. Successful traders have a larger edge and better money management than unsuccessful traders.
Unlike popular belief however, this study shows that the smaller edge of unsuccessful traders is not the cause of their failure. Traders failure can be explained almost exclusively by their poor money management practices.
The techniques taught in this book will give you the trading edge you need. If you want to increase this edge, the importance of money management cannot be stressed enough.
Appendix
HISTORICAL VOLATILITY CALCULATIONS
The historical volatility is defined as the standard deviation of the logarithmic price changes measured at regular intervals of time. Since settlement prices are usually considered the most reliable, the most common method of computing volatility involves using settlement-tosettlement price changes. We defined each price change, Xi, as:
204
We first calculate the standard deviation of the logarithmic price changes:
We then calculate the annual volatility by multiplying the standard deviation by the square root of the time interval between price changes. Since we looked at price changes every week, the time interval is 365/7:
Repented from: Nathanberg Sheldon. Option Volatility & Pricing, Advanced Trading Strategies and lechniques, 2d ed, (Chicago: Probus Publishing, 1994), Appendix B.
205
MOORE RESEARCH CENTER-STATISTICAL STUDIES
Please read this paragraph before examining the following tables! We wish to be perfectly clear that we are not testing mechanical systems. Rather, we are examining a variable to see if there is a tendency that might be useful as part of an entry or exit methodology. Your eye might initially be drawn to the column reading "average net." This is not meant to imply any level of profitability. We are more concerned with the percentages to see if a market goes up or down more in response to an initial condition. We are also testing to see if an indicator has uniform behaviour across all markets.
The following text will help you better understand how to read the tables.
Introduction
Moore Research Center of Eugene, Oregon, has provided statistical testing to supplement our own research efforts. These studies highlight certain market tendencies and serve as a way to quantify characteristics of market behaviour. The tests also provide such information as frequency of pattern occurrence, directional bias, and daily bar characteristics. We use this statistical testing for its comparative value only. It is not meant to represent a mechanical system. Thus, no statistics such as commissions or slippage are factored in, nor are data on total profitability or maximum drawdown provided.
We will present a short description of each test and comment on the results. Before we continue, though, it is important to mention the testing methodology. The tests are run on actual contract data which are rolled from the lead contract to the next front month either one day before first notice day, or five days before expiration (whichever comes first). We run the buy and sell tests independently to examine the potential for directional bias.
Since we are looking for tendencies or probabilities, sample size is important. The tests are run on 10 years of data across 25 markets. The number of days examined is listed in the column labelled "Total Days." The total number of occurrences tested can be calculated by dividing the number of wins by the percentage of wins. For example, 48(60%) indicates that there were 48 positive results with a "win/loss"
ratio of 60%. 48/.6 = 80 total occurrences.
206
We also like to see if a pattern or relationship holds true across multiple markets.
This builds our confidence that the pattern represents a true principle of market behaviour. Most of the studies are run on just one to two variables. We feel that these results are robust and that the relationships should hold true in the future. We also hope that you can use these tests as a departure point in your own system development or examination of market behaviour.
Daily Bar Statistical Profiles
This first set of studies profiles a market's price action following a day that has closed at the extreme of its range. The tests show a tendency for an intraday reversal following this condition. This pattern forms a basis for the 80-20's bar strategy discussed in Chapter 6.
We will describe each column on the studies so that you can examine the results for yourself. First is a column listing the total number of days tested. This is followed by a listing titled "Setup Days." Here, the frequency of occurrence is shown. The next column profiles the percentage of time that the market opened up or down.
After this, we can see the number of times that the market penetrated the previous day's high or low and the average amount of the penetration. Finally, the study shows the percentage of time that the market closed up or down.
Let's look at the first analysis titled "Historical Upper 90%." This study profiles the market's action following a day where the market closed in the upper 10 percent of its range. If we look at the S&Ps, we can see that this setup occurred 17 percent of the time across 2436 days. The next morning the market opened up 48 percent of the time. It then penetrated the previous day's high 85 percent of the time by an average amount of 2.00 points. Lastly, it closed up only 50 percent of the time.
If we look at the next study, it profiles the market's action following days closing in the upper 20 percent of the range. As you can see, the results are just about the same. This same tendency towards mid-day reversal also holds true for markets closing in the lower extreme of their range in fact, even more so. When the S&Ps closed in the lower 20 percent of their range, there was only a 42 percent probability that they would close lower the following day
It is also interesting to note that despite the upside bias to the S&Ps and the bonds over the last 10 years, the penetrations through the previous days extremes tended to be significantly deeper on the sell side.
207
The last set of studies in this section profiles the day following a WR7, which is the widest range of the last seven days. We examined both WR7s with a high close and WR7s with a low close. The probabilities for the majority of markets favoured a close in the opposite direction the following day Sugar was an extreme example. It closed in the opposite direction 60 percent of the time. These tests would indicate that it is definitely a good idea to take profits on a range expansion bar (in other words, one with a very wide range). They also suggest that the WR7 day could be a useful pattern to combine with 80-20's bars.
208
EXHIBIT A.1 Historical Upper 90% Statistics
209
EXHIBIT A.2 Historical Upper 80% Statistics
210
EXHIBIT A.3 Historical Lower 10% Statistics
211
EXHIBIT A.4 Historical Lower 20% Statistics
212
EXHIBIT A.5 Historical WR7 and Higher-Close Statistics
213
EXHIBIT A.6 Historical WR7 and Lower-Close Statistics
214 Historical ROC Reports
This study examines the indicator used for part of the momentum pinball strategy. A three-period RSI is run on a one-period rate of change. If the reading is greater than 70, the next day's opening is sold. If the reading is less than 30. the next day's opening is bought. We then examine the initial entry at the following morning's open and the following day's close. No protective stops are used. Our main intent is to examine the difference between the setup day's opening and the following day's opening. We are looking for evidence to support our theory that this ROC/RSI indicator has some value as a short-term overbought/ oversold study to help capture Taylor's swing trading rhythm. We definitely feel that there is enough of a tendency displayed by this indicator that other people might be interested in examining it for their own system development.
This set of studies is run off daily bars only. It is not meant to be a representation of the complete momentum pinball strategy that employs a first hour range breakout.